HOLO Audio Spring DAC - R2R DSD512

[ted_b]

31 minutes ago, Miska said:

I wouldn't use NOS mode for any rate lower than 352.8k. It will generate a lot of high frequency hash/images/distortion.

Jussi, I am a bit confused by this comment.  In NOS mode the dac is slave to HQPlayer and its output.  Are you saying HQPlayer 44k to, say, 192k would sound that way??  In all fairness, I upsample to 352/384k but not everyone might.

[westelk]

19 minutes ago, ted_b said:

Jussi, I am a bit confused by this comment.  In NOS mode the dac is slave to HQPlayer and its output.  Are you saying HQPlayer 44k to, say, 192k would sound that way??  In all fairness, I upsample to 352/384k but not everyone might.

Miska,

What mode do you suggest for pcm up to 192k?

[Miska]

3 minutes ago, westelk said:

What mode do you suggest for pcm up to 192k?

 

Either use NOS mode and upsample in playback software like I do. Or alternatively use the "OS" or "OS PCM" mode. The "OS DSD" mode is not worth using, the algorithm is too poor.

 

[Miska]

30 minutes ago, ted_b said:

Jussi, I am a bit confused by this comment.  In NOS mode the dac is slave to HQPlayer and its output.  Are you saying HQPlayer 44k to, say, 192k would sound that way??  In all fairness, I upsample to 352/384k but not everyone might.

 

Yeah, if someone uses HQPlayer (or A+ or something else) to upsample to 352.8/384k and/or DSD, then NOS mode is what should be used.

 

If someone doesn't do upsampling in software, or sends in PCM at rate lower than 352.8/384k, then NOS mode shouldn't be used. The PCM -> PCM upsampling in the box is quite OK, the DSD one is not...

 

[elcorso]

3 hours ago, Miska said:

 

DSD output is 6 dB lower than PCM output. Thus the output voltage is half. Which is expected, because DSD sources sometimes exceed 0 dB level. So the technical absolute maximums match.

 

I wouldn't use NOS mode for any rate lower than 352.8k. It will generate a lot of high frequency hash/images/distortion.

 

Thanks for the explanation. When I mentioned NOS was with respect to the button in the DAC, I always do it by means of software, HQ Player or A +, but with this DAC I sometimes find excellent recordings that practically do not require upsampling. Not many, but there are.

 

The 6dB difference between PCM and DSD is well known to DSD lovers (like me), but I find the explanation in the published specifications for this DAC somewhat confusing.

 

Best,

 

Roch

[guymrob]

6 hours ago, Miska said:

I wouldn't use NOS mode for any rate lower than 352.8k. It will generate a lot of high frequency hash/images/distortion.

 

I do agreed but the whole thing about having NOS mode is to bypass the over-sampling digital filter which in turned create an almost perfect impulse response (virtually no ringing). This gives Spring its unique 'organic' and real sounding character. Though noise will be an issue, it will get some degree of attenuation via analogue LPF inside Spring, to the amp and speakers.

 

In my listening test, I always prefer listening in NOS even in 44.1k non-upsampled. I don't hear any form of distortion, they all sound very 'real' to me.

[Miska]

6 hours ago, guymrob said:

I do agreed but the whole thing about having NOS mode is to bypass the over-sampling digital filter which in turned create an almost perfect impulse response (virtually no ringing). This gives Spring its unique 'organic' and real sounding character. Though noise will be an issue, it will get some degree of attenuation via analogue LPF inside Spring, to the amp and speakers.

 

For proper reconstruction of RedBook, the analog filter would need to have 96 dB attenuation by 24.1 kHz, but nothing yet at 20 kHz. So 96 dB attenuation in 4.1 kHz wide band. That won't happen, not even nearly. Only practical way to do it is a digital filter.

 

Another thing is that the "ringing" is originally created when the recording is created. Either at ADC when if it is running at 44.1 kHz, or at later time when for example 96 kHz master is converted to 44.1 kHz for distribution. You need to have an apodizing upsampling filter to later change/reduce it.

 

6 hours ago, guymrob said:

In my listening test, I always prefer listening in NOS even in 44.1k non-upsampled. I don't hear any form of distortion, they all sound very 'real' to me.

 

That way you'll also get 3 dB HF roll-off... And quite a bit of timing uncertainty for transients, depending on where sample happened to be in regards to transient timing.

 

This is how 19 kHz sine wave looks at 44.1 kHz sampling rate from a NOS DAC:

 

And same after upsampling to 384k sampling rate:

 

Makes quite a bit of difference?

 

P.S. If you have recordings in DXD or DSD, you can play it through as-is and it'll come out just fine... And you don't have any of the problems compared to reproduction of RedBook.

[chrille]

That looks interesting.

But still not a perfect sinewave.

Would the same apply with all DACs even Chord DACs?

Rob Watts advices against software upsampling with his DACs.

Would there be a measurable improvement of sinewaves with his DACs with 16/44.1 upsampled or not?

If I understand things correctly he only recommends  his own M-scaler upsampler.

What about Benchmark's DAC 2 and 3?

If you have answers to my questions please respond.

[Miska]

4 hours ago, chrille said:

But still not a perfect sinewave.

 

What is not perfect in the sinewave there? There are no visible distortions anymore... You can also see that each span overlays perfectly with red color, there are no more those blue regions indicating variations.

 

4 hours ago, chrille said:

Would the same apply with all DACs even Chord DACs?

 

You cannot run Chord DACs in NOS mode, thus no.

 

4 hours ago, chrille said:

Rob Watts advices against software upsampling with his DACs.

 

That is completely unrelated. And off-topic for this thread.

 

4 hours ago, chrille said:

Would there be a measurable improvement of sinewaves with his DACs with 16/44.1 upsampled or not?

 

To some degree maybe.

 

4 hours ago, chrille said:

What about Benchmark's DAC 2 and 3?

 

Same thing as with Chord DACs, there's no possibility to run in NOS mode. So you'll always get upsampling, internal or external.

[pkane2001]

18 hours ago, Miska said:

Yeah, if someone uses HQPlayer (or A+ or something else) to upsample to 352.8/384k and/or DSD, then NOS mode is what should be used.

 

Jussi, so what's the reason to use HQPlayer upsampling/filtering and then Holo Spring in NOS mode? Is it that HQPlayer has a better implementation? Higher precision? More filter options? What's wrong with Holo Spring filters or upscaling algorithm? 

 

I have mine configured as you recommend: HQPlayer converting all content to DSD512, using xtr-mp-2s filters,  HoloSpring in NOS mode.

[ted_b]

Yes, HQPlayer's filters and modulators are 10x better than most dacs, let alone Holo's.  One of Holo's biggest strengths is being able to bypass their filters...i.e NOS mode.  Many dacs do not have that setting.

[barrows]

But, although many DACs do not have a "NOS" mode, if you feed DSD 128, 256, 512, or PCM 352.8/384 to most DACs, what will happen is a "bypass" of the first filter stage.  You do not need a switchable "NOS" mode to take advantage of external oversampling, or to defeat the first filter stage in most DACs-just by feeding a DAC an already oversampled signal you will bypass the first onboard filter stage.

Now in many DACs there will be a second oversampling stage, which will not be bypassed by feeding 352.8/384 PCM, but this stage, if well implemented at all, will have virtually no audible consequences (as all oversampling artifacts at this high a rate of oversampling will be in such high frequencies as to inaudible).

[pkane2001]

57 minutes ago, ted_b said:

Yes, HQPlayer's filters and modulators are 10x better than most dacs, let alone Holo's.  One of Holo's biggest strengths is being able to bypass their filters...i.e NOS mode.  Many dacs do not have that setting.

 

10x better in what sense? I understand the theoretical benefit, but what's the practical one? I certainly don't hear a 10x better sound quality in NOS mode 

[westelk]

Several people have mentioned using the ISO Regen with the Holo dac, how about the Sotm tx-USBultra? My current source is an Aurender N100 feeding the Holo USB input. I also have a Singxer SU-1 available but not currently using it. Has anyone compared the ISO Regen against the SOtM?

[elcorso]

1 minute ago, pkane2001 said:

 

10x better in what sense? I understand the theoretical benefit, but what's the practical one? I certainly don't hear a 10x better sound quality in NOS mode 

 

I feel the same as Ted, and also playing the Holo by A+....

 

But this is in the very individual taste territory, where nobody is wrong, nobody is right, but happy 

 

Roch

[elcorso]

7 hours ago, Miska said:

 

For proper reconstruction of RedBook, the analog filter would need to have 96 dB attenuation by 24.1 kHz, but nothing yet at 20 kHz. So 96 dB attenuation in 4.1 kHz wide band. That won't happen, not even nearly. Only practical way to do it is a digital filter.

 

Another thing is that the "ringing" is originally created when the recording is created. Either at ADC when if it is running at 44.1 kHz, or at later time when for example 96 kHz master is converted to 44.1 kHz for distribution. You need to have an apodizing upsampling filter to later change/reduce it.

 

 

That way you'll also get 3 dB HF roll-off... And quite a bit of timing uncertainty for transients, depending on where sample happened to be in regards to transient timing.

 

This is how 19 kHz sine wave looks at 44.1 kHz sampling rate from a NOS DAC:

 

And same after upsampling to 384k sampling rate:

 

Makes quite a bit of difference?

 

P.S. If you have recordings in DXD or DSD, you can play it through as-is and it'll come out just fine... And you don't have any of the problems compared to reproduction of RedBook.

 

Fully agree with you !

 

But finding a lot of well-recorded music still persists 

 

Roch

[elcorso]

29 minutes ago, barrows said:

But, although many DACs do not have a "NOS" mode, if you feed DSD 128, 256, 512, or PCM 352.8/384 to most DACs, what will happen is a "bypass" of the first filter stage.  You do not need a switchable "NOS" mode to take advantage of external oversampling, or to defeat the first filter stage in most DACs-just by feeding a DAC an already oversampled signal you will bypass the first onboard filter stage.

Now in many DACs there will be a second oversampling stage, which will not be bypassed by feeding 352.8/384 PCM, but this stage, if well implemented at all, will have virtually no audible consequences (as all oversampling artifacts at this high a rate of oversampling will be in such high frequencies as to inaudible).

 

Maybe the problem is that most of the music that I have and I like is in 16/44 ...!

 

Roch

[ted_b]

30 minutes ago, barrows said:

But, although many DACs do not have a "NOS" mode, if you feed DSD 128, 256, 512, or PCM 352.8/384 to most DACs, what will happen is a "bypass" of the first filter stage.  You do not need a switchable "NOS" mode to take advantage of external oversampling, or to defeat the first filter stage in most DACs-just by feeding a DAC an already oversampled signal you will bypass the first onboard filter stage.

Now in many DACs there will be a second oversampling stage, which will not be bypassed by feeding 352.8/384 PCM, but this stage, if well implemented at all, will have virtually no audible consequences (as all oversampling artifacts at this high a rate of oversampling will be in such high frequencies as to inaudible).

Agree, and its nice to find dacs that have those excellent second stages....but we are getting far afield from the question asked about NOS in the Holo.  Pkane2001 specifically asked about why NOS in the Holo.  Jussi has already said that the PCM filters aren't bad, but DSD stuff is average at best (bias notwithstanding ).  

 

Pkane2001, I think software upsampling vs hardware upsampling has been discussed ad nauseum.   I find it to be the same for a lot of "integrated vs separates" logic; the "separates" (in this case software upsampling using powerful cpus) are built for the task, the "integrateds" have to be jack of all trades, given a price point.  In some cases, the integrateds have no bypass or coexist capability..in the case of dacs like Holo they do.   Oversimplified, sorry.

[pkane2001]

3 minutes ago, ted_b said:

Pkane2001, I think software upsampling vs hardware upsampling has been discussed ad nauseum.   I find it to be the same for a lot of "integrated vs separates" logic; the "separates" (in this case software upsampling using powerful cpus) are built for the task, the "integrateds" have to be jack of all trades, given a price point.  In some cases, the integrateds have no bypass or coexist capability..in the case of dacs like Holo they do.   Oversimplified, sorry.

 

Thanks, Ted, I get the general appeal of the software-based approach. I was curious about the details as to what @Miska found to be inferior with Holo Springs filters compared to HQPlayer.

 

[Miska]

1 hour ago, barrows said:

Now in many DACs there will be a second oversampling stage, which will not be bypassed by feeding 352.8/384 PCM, but this stage, if well implemented at all, will have virtually no audible consequences (as all oversampling artifacts at this high a rate of oversampling will be in such high frequencies as to inaudible).

 

That second stage is typically sample-and-hold (aka S/H, SAH, ZOH), so not really doing anything but copying same sample N-times. Plus if it's a delta-sigma DAC it doesn't bypass the modulator. Sometimes the second stage is linear interpolation which is not much better than S/H.

 

So what you see in the output is images around multiples of 352.8 kHz. Those are absent when for example properly upsampled DSD input is used...

 

[Miska]

Here's wideband output spectrum 0 - 22.05 kHz sweep with 44.1/32 source.

 

Upsampled to 352.8k PCM using built-in chip:

 

Upsampled to 352.8k PCM using HQPlayer:

 

Upsampled to DSD using built-in chip:

 

Upsampled to DSD512 using HQPlayer:

 

 

You can see that with PCM there are about three images still visible around multiples of the 352.8k sampling rate, not completely eliminated by the analog filter (completely typical amount, no surprises).

 

With DSD512 from HQPlayer you can see that there are no images or other extra visible from the digital source, just analog noises. There are no images, because HQPlayer runs digital filter up to 22.5792 MHz rate (real 512x filter), and multiples of that rate are completely eliminated by the analog filter.

 

 

Someone also asked earlier about the Jtest-24 results for the DAC's built in USB interface:

Cannot really get much better than that easily...

 

[hifi25nl]

Thanks, Miska. The result of Jtest-24 is impressive and confirm my listening tests! 

[pkane2001]

37 minutes ago, Miska said:

Here's wideband output spectrum 0 - 22.05 kHz sweep with 44.1/32 source.

 

Upsampled to 352.8k PCM using built-in chip:

 

Upsampled to 352.8k PCM using HQPlayer:

 

Upsampled to DSD using built-in chip:

 

Upsampled to DSD512 using HQPlayer:

 

 

You can see that with PCM there are about three images still visible around multiples of the 352.8k sampling rate, not completely eliminated by the analog filter (completely typical amount, no surprises).

 

With DSD512 from HQPlayer you can see that there are no images or other extra visible from the digital source, just analog noises. There are no images, because HQPlayer runs digital filter up to 22.5792 MHz rate (real 512x filter), and multiples of that rate are completely eliminated by the analog filter.

 

 

Someone also asked earlier about the Jtest-24 results for the DAC's built in USB interface:

Cannot really get much better than that easily...

 

 

That is really excellent performance, Jussi. Thanks for sharing these!

 

[chsas75]

Audiophonics sales Holo Spring with their logo with USB on XMOS 208 and halvanic isolation.

 

 

 

ARMATURE ASTERION

 

What is this plate producer good way to upgrade the usb in.

[rayl1234]

10 hours ago, Miska said:

For proper reconstruction of RedBook, the analog filter would need to have 96 dB attenuation by 24.1 kHz, but nothing yet at 20 kHz. So 96 dB attenuation in 4.1 kHz wide band. That won't happen, not even nearly. Only practical way to do it is a digital filter.

 

I agree with everything but the very last sentence. Once you have aliasing due to ADC if a signal that is not sufficiently bandwidth limited (bec as stated, no filter can be 96dB down in 2kHz for 44.1 or in 4kHz for 48), that aliasing loss is irrecoverable*.

 

No digital filter after the aliased pcm is output by the ADC can correct that. We can thank Nyquist for demonstrating this to us well before Red Book was invented. 

 

If we are saying we ADC at a higher sampling rate and want to low pass in digital, then yes, digital filters can be steeper...

 

 

 

*Except maybe for some fancy source model like modeling the instruments, voices, etc., which doesn’t exist aside from spy gear aimed at picking out cocktail conversations from afar...